Name
Abstract | ||
|---|---|---|
Invariance to photometric changes is implicitly required for a view-based object recognition sys- The definition of reliable local signal characteri- tem. zations is of great importance for many computer vision tasks as mosai'cing, 3D-scene reconstruction or more recently in applications like content-based image retrieval systems. The following study con- cerns this last general pattern. Aiming at this, we present the use of Full-Zernike moments as a local characterization of the image signal. Their compu- tation allows us to construct an invariant vector, of which the projection in an index table (feature space) provides a vote for some model-images. This approach is based on the quasi-invariant theory ap- plied to perspective transformations and is an exten- sion of a standard point to point matching between two images. It addresses the problem of similarity search in high dimensional space (d > 20). In this article we propose the use of Zernike mo- ments as a local description of feature points. We describe the so-computed quasi-invariant vector in section 2. A particular attention will be devoted to the invariance against rotation that is achieved with- out loss of the completeness properties of the set. In section 3 we present an adapted treatment in order to obtain the invariance against large scale changes (> 20%) regarding the scale-space theory. Further- more, a normalization of the signal carries out an in- variance against locally affine photometric changes. We have evaluated the capabilities of the proposed description for a simple matching task, and for im- agelobject retrieval. In the last section, we describe the first results obtained with the use of an ori~inal - clustering sheme (14) in order to avoid an exhaustive scanning of the database. |
Year | Venue | Keywords |
|---|---|---|
2000 | MVA | similarity search,computer vision,scale space,indexation,invariant theory,object recognition,feature space,point to point |
Field | DocType | Citations |
Computer vision,Feature vector,Pattern recognition,Scale space,Image retrieval,Zernike polynomials,Artificial intelligence,Invariant (mathematics),Velocity Moments,Nearest neighbor search,Mathematics,Cognitive neuroscience of visual object recognition | Conference | 2 |
PageRank | References | Authors |
0.39 | 8 | 3 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Erwan Bigorgne | 1 | 21 | 3.89 |
| Catherine Achard | 2 | 158 | 19.60 |
| Jean Devars | 3 | 31 | 5.61 |